Research on Black-Box Modeling Prediction of USV Maneuvering Based on SSA-WLS-SVM

Round 1

Reviewer 1 Report

The paper concerns with black box motion identification of USV maneuvering by mean of SVM algorithm. The topic is interesting and the authors presented their study very clearly. My personal opinion is that the paper can be published on the journal after very minor revision. 

1) The authors should better stress the fact that black box modeling is the best choice (with respect to white box modeling) for automatic control purposes. On the other hand, white box is probably superior for the improvement of the hydrodynamic design of the vehicle. 

2) In the introduction: in the introductory chapter, specifically in the discussion of white box modeling, the authors do not consider the fact that white box paradigm can help to improve the original mathematical model. For example, Bonci et al, (Methods for estimating parameters of practical ship maneuvering model by the combination of RANSE computations and System Identification, Applied Ocean Research, 2015, vol 52, p.274-294), improved the Abkovitz model (fully non linear) by identification of CFD free running simulations of a USV model. This work, used high accurate CFD data with the aim to remove the noise that typically characterizes experimental data in outdoor basins. This aspect can be briefly included in the white box modeling. 

3) When considering the maneuvers obtained by MMG model, was random noise added to the output variables of the kinematics (trajectory, speed and acceleration?). Could this element have a potential worsening effect on the training process of SVM? In this regard, use of CFD data of free running maneuvers (high accurate and free of undesired noise) can improve the prediction capability of the algorithm? 

Thank you

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Thank you for this paper on USV motion prediction using SVM. I have a few comments on the paper, some minor ones and some major ones.

Major

I am not quite sure what you ultimate goal is. From your description I gather that you want to predict the motion of the USV. To what purpose? Do you want to compensate for disturbances in order to achieve an as accurate actual motion as possible?

What are the actual inputs and outputs of your SVM? I suggest providing a table with all the relevant data. Are the inputs measurable in the real world? From the paper I read hydrodynamic forces. However, one usually measured pressures on the hull etc. So I am not sure how the SVM based on the simulation would transform into the real world. Please clarify.

You provide almost no information on the CFD simulation (mesh, fluid model, ...). Hence, one cannot replicate and evaluate your work. Please provide lots of details here. Some images of the results - in order to evaluate significance of the parameters used - would also be very helpful.

Minor

Lines 67 - 73 should be deleted

Figure 5 and line 74: I think it should read "speed" and not "velocity" here.

Figure 6: I think (a) and (b) are interchanged.

Figures in general: You use dashed red lines throughout. However, in many cases the separation of the dashes is too coarse (for example fig. 8(a)). I would suggest using finely dotted lines instead.

Overall

Due to the major comments you would suggest a major revision. However, I must also point out that, while reading your paper, I was very much reminded of work done on AUVs about 10 years ago. I would suggest looking into a combination of the keywords CFD, machine learning, SVM, AUVs, flow prediction, motion prediction.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Thank you for the updated manuscript. I have no further comments.